Identifying Matching Transfer Transactions

1. A method, comprising: obtaining a first plurality of unpaired transactions, each comprising a transaction ID, an entity ID, and a plurality of attributes that each correspond to an attribute type of one or more attribute types; deriving a plurality of transaction groups by grouping the first plurality of unpaired transactions using their entity IDs; dividing a transaction group of the plurality of transaction groups into a first transaction subgroup and a second transaction subgroup; constructing a tree for the first transaction subgroup based on a first attribute type of the one or more attribute types; training, using distances generated from attributes of the first attribute type in pairs of historical transactions, a machine learning model to set a distance constraint corresponding to the first attribute type, wherein training the machine learning model comprises: obtaining the pairs of historical transactions, wherein each pair corresponds to a label indicating whether transaction IDs of the pair are comprised by a matched transfer pair, and calculating a distance for the pair using a distance measure of a matching criterion corresponding to the first attribute type; matching, using the machine learning model, a first transaction of the first transaction subgroup with a second transaction of the second transaction subgroup by searching the tree while applying a first matching criterion to the transactions of the second transaction subgroup, the first matching criterion corresponding to the first attribute type, wherein applying the first matching criterion comprises determining whether a distance between a first attribute of the first transaction and a second attribute of the second transaction satisfies the distance constraint, and wherein the first attribute of the first transaction and the second attribute of the second transaction correspond to the first attribute type; and in response to matching the first transaction with the second transaction, forming a first matched transfer pair comprising the entity ID of the transaction group, the transaction ID of the first transaction, and the transaction ID of the second transaction.

2. The method of claim 1 , further comprising: verifying that the transaction group satisfies one or more verification constraints corresponding to one or more of the plurality of attributes, wherein the transaction group is divided into the first transaction subgroup and the second transaction subgroup in response to the verifying.

3. The method of claim 1 , wherein the first matched transfer pair is the only matched transfer pair comprising the transaction ID of the first transaction, the transaction ID of the second transaction, and the entity ID of the transaction group.

4. The method of claim 1 , further comprising: calculating a metric corresponding to the entity ID of the transaction group by excluding the first matched transfer pair from a subset of the first plurality of unpaired transactions that comprise the entity ID of the transaction group.

5. The method of claim 1 , further comprising: receiving a second plurality of transactions; determining that the second plurality of transactions updates the first transaction; matching the updated first transaction with a third transaction of the second transaction subgroup by applying the first matching criterion to the transactions of the second transaction subgroup; and in response to matching the updated first transaction with the third transaction, replacing, in the first matched transfer pair, the transaction ID of the second transaction with the transaction ID of the third transaction.

6. The method of claim 1 , further comprising: receiving a second plurality of transactions comprising a third transaction, the third transaction comprising the entity ID of the transaction group; assigning the third transaction to the first transaction subgroup; matching the third transaction with a fourth transaction of the second transaction subgroup by applying the first matching criterion to the transactions of the second transaction subgroup; and in response to matching the third transaction with the fourth transaction, forming a second matched transfer pair comprising the entity ID of the transaction group, the transaction ID of the third transaction, and the transaction ID of the fourth transaction.

7. The method of claim 1 , wherein matching the first transaction with the second transaction further comprises searching the tree while applying a second matching criterion corresponding to a second attribute type of the one or more attribute types, the method further comprising: combining, using a first attribute relevance factor corresponding to the first attribute type and a second attribute relevance factor corresponding to the second attribute type, a first distance corresponding to the first attribute and a second distance corresponding to the second attribute; and setting, using the machine learning model, the first attribute relevance factor and the second attribute relevance factor.

8. A system, comprising: a memory coupled to a computer processor; a repository configured to store: a first plurality of unpaired transactions, each comprising a transaction ID, an entity ID, and a plurality of attributes that each correspond to an attribute type of one or more attribute types, a plurality of transaction groups, and a tree for a first transaction subgroup of a transaction group of the plurality of transaction groups, wherein the transaction group corresponds to the entity ID; and a transaction analyzer, executing on the computer processor and using the memory, configured to: derive the plurality of transaction groups by grouping the first plurality of unpaired transactions using their entity IDs; divide the transaction group into the first transaction subgroup and a second transaction subgroup; construct the tree for the first transaction subgroup based on a first attribute type of the one or more attribute types; train, using distances generated from attributes of the first attribute type in pairs of historical transactions, a machine learning model to set a distance constraint corresponding to the first attribute type, wherein training the machine learning model comprises: obtaining the pairs of historical transactions, wherein each pair corresponds to a label indicating whether transaction IDs of the pair are comprised by a matched transfer pair, and calculating a distance for the pair using a distance measure of a matching criterion corresponding to the first attribute type; match, using the machine learning model, a first transaction of the first transaction subgroup with a second transaction of the second transaction subgroup by searching the tree while applying a first matching criterion to the transactions of the second transaction subgroup, the first matching criterion corresponding to the first attribute type, wherein applying the first matching criterion comprises determining whether a distance between a first attribute of the first transaction and a second attribute of the second transaction satisfies the distance constraint, and wherein the first attribute of the first transaction and the second attribute of the second transaction correspond to the first attribute type; and in response to matching the first transaction with the second transaction, form a first matched transfer pair comprising the entity ID of the transaction group, the transaction ID of the first transaction, and the transaction ID of the second transaction.

9. The system of claim 8 , wherein the transaction analyzer is further configured to: verify that the transaction group satisfies one or more verification constraints corresponding to one or more of the plurality of attributes, wherein the transaction group is divided into the first transaction subgroup and the second transaction subgroup in response to the verifying.

10. The system of claim 8 , further comprising an application executing on the computer processor and using the memory, configured to: calculate a metric corresponding to the entity ID of the transaction group by excluding the first matched transfer pair from a subset of the first plurality of unpaired transactions that comprise the entity ID of the transaction group.

11. The system of claim 8 , wherein the transaction analyzer is further configured to: receive a second plurality of transactions; determine that the second plurality of transactions updates the first transaction; match the updated first transaction with a third transaction of the second transaction subgroup by applying the first matching criterion to the transactions of the second transaction subgroup; and in response to matching the updated first transaction with the third transaction, replace, in the first matched transfer pair, the transaction ID of the second transaction with the transaction ID of the third transaction.

12. The system of claim 8 , wherein the transaction analyzer is further configured to: receive a second plurality of transactions comprising a third transaction, the third transaction comprising the entity ID of the transaction group; assign the third transaction to the first transaction subgroup; match the third transaction with a fourth transaction of the second transaction subgroup by applying the first matching criterion to the transactions of the second transaction subgroup; and in response to matching the third transaction with the fourth transaction, form a second matched transfer pair comprising the entity ID of the transaction group, the transaction ID of the third transaction, and the transaction ID of the fourth transaction.

13. A method, comprising: obtaining a first plurality of unpaired transactions comprising a transaction ID, a business entity ID, and a timestamp; deriving a plurality of transaction groups by grouping the first plurality of unpaired transactions using their business entity IDs; dividing a transaction group of the plurality of transaction groups into a first transaction subgroup and a second transaction subgroup, the transaction group corresponding to a first business entity ID; constructing a tree for the first transaction subgroup based on the timestamp; training, using distances generated from timestamps in pairs of historical transactions, a machine learning model to set a threshold time interval, wherein training the machine learning model comprises: obtaining the pairs of historical transactions, wherein each pair corresponds to a label indicating whether transaction IDs of the pair are comprised by a matched transfer pair and calculating a distance for the pair using a distance measure of a matching criterion corresponding to the timestamp; matching, using the machine learning model, a first transaction of the first transaction subgroup with a second transaction of the second transaction subgroup by searching the tree while applying a first matching criterion to the transactions of the second transaction subgroup, the first matching criterion corresponding to the timestamp, wherein applying the first matching criterion comprises determining whether a distance between a first timestamp of the first transaction and a second timestamp of the second transaction satisfies the threshold time interval; and in response to matching the first transaction with the second transaction, forming a first matched transfer pair comprising the first business entity ID, the transaction ID of the first transaction, and the transaction ID of the second transaction.

14. The method of claim 13 , wherein the first plurality of unpaired transactions further comprises an account ID and an amount, the method further comprising: verifying that the transaction group comprises: a transaction comprising an amount with a positive sign, a transaction comprising an amount with a negative sign, and two transactions comprising different account IDs, wherein the transaction group is divided into the first transaction subgroup and the second transaction subgroup in response to the verifying.

15. The method of claim 13 , further comprising: calculating a financial metric for the first business entity ID by excluding the first matched transfer pair from a subset of the first plurality of unpaired transactions that comprise the first business entity ID; and making a lending decision using the financial metric for the first business entity ID.

16. The method of claim 13 , further comprising: receiving a second plurality of transactions comprising a third transaction, the third transaction comprising the first business entity ID; assigning the third transaction to the first transaction subgroup; matching the third transaction with a fourth transaction of the second transaction subgroup by applying the first matching criterion to the transactions of the second transaction subgroup; and in response to matching the third transaction with the fourth transaction, forming a second matched transfer pair comprising the first business entity ID, the transaction ID of the third transaction, and the transaction ID of the fourth transaction.

17. The method of claim 13 , wherein the first plurality of unpaired transactions further comprises an amount, wherein matching the first transaction with the second transaction further comprises searching the tree while applying a second matching criterion to the amounts of the transactions of the second transaction subgroup, the method further comprising: combining, using a first attribute relevance factor corresponding to the timestamp and a second attribute relevance factor corresponding to the amount, a first distance corresponding to the timestamp and a second distance corresponding to the amount; and setting, using the machine learning model, the first attribute relevance factor and the second attribute relevance factor.